diff --git a/scarf/readers.py b/scarf/readers.py
index 148059f..e1618c7 100644
--- a/scarf/readers.py
+++ b/scarf/readers.py
@@ -12,7 +12,7 @@
 import math
 import os
 from abc import ABC, abstractmethod
-from typing import IO, Dict, Generator, List, Optional, Tuple
+from typing import IO, Dict, Generator, List, Optional, Tuple, Union
 
 import h5py
 import numpy as np
@@ -769,10 +769,7 @@ def feat_ids(self) -> np.ndarray:
     def feat_names(self) -> np.ndarray:
         """Returns a list of feature names."""
         if self._check_exists(self.featureAttrsKey, self.featNamesKey):
-            if self.groupCodes[self.featureAttrsKey] == 1:
-                values = self.h5[self.featureAttrsKey][self.featNamesKey]
-            else:
-                values = self.h5[self.featureAttrsKey][self.featNamesKey][:]
+            values = self.h5[self.featureAttrsKey][self.featNamesKey]
             return self._replace_category_values(
                 values, self.featNamesKey, self.featureAttrsKey
             ).astype(object)
@@ -781,7 +778,28 @@ def feat_names(self) -> np.ndarray:
         )
         return self.feat_ids()
 
-    def _replace_category_values(self, v: np.ndarray, key: str, group: str):
+    def _replace_category_values(
+        self, v: Union[np.ndarray, h5py.Group, h5py.Dataset], key: str, group: str
+    ) -> np.ndarray:
+        # check if v is a Group with codes + categories structure
+        if isinstance(v, h5py.Group):
+            if "codes" in v and "categories" in v:
+                codes = v["codes"][:]
+                categories = v["categories"][:]
+                try:
+                    return np.array([categories[x] for x in codes])
+                except (IndexError, TypeError):
+                    logger.warning(f"Failed to decode categorical data for {key}")
+                    return np.array([f"feature_{x}" for x in range(len(codes))])
+            else:
+                # It's a Group but doesn't have the expected structure, try to read it as dataset
+                logger.warning(f"{key} is a Group but missing 'codes' or 'categories', attempting to extract data")
+                return v[:]
+        
+        # if v is a Dataset
+        if isinstance(v, h5py.Dataset):
+            v = v[:]
+        
         if self.catNamesKey is not None:
             if self._check_exists(group, self.catNamesKey):
                 cat_g = self.h5[group][self.catNamesKey]